Skip to content

Skill: claim ledger, four loops, mandatory adversarial review - #5

Merged
kgraph57 merged 1 commit into
mainfrom
feat/skill-verification-loops
Aug 13, 2026
Merged

Skill: claim ledger, four loops, mandatory adversarial review#5
kgraph57 merged 1 commit into
mainfrom
feat/skill-verification-loops

Conversation

@kgraph57

Copy link
Copy Markdown
Owner

The medical-fact-check skill is no longer a linear 15-item checklist.

Pipeline: acquire → claim ledger → engine (existence) → semantic honesty → 15-criteria → adversarial red-team → score → correction loop.

Four named loops (cap 3 on adversarial/correction): engine retry once then unresolved (never guess Hallucination); semantic one extra abstract lookup per T1 cite; adversarial KILL/MAJOR/MINOR/PASS; correction re-enters from the engine.

KILL (Tier 4 presented as real, or harm) forces score ≤ D. MAJOR cannot be A. Not CDS.

Add a claim ledger, engine/semantic/adversarial/correction loops (cap 3), and a mandatory red-team (KILL/MAJOR/MINOR/PASS) so the skill cannot score A while fabricated citations or unsafe advice remain.
@kgraph57
kgraph57 merged commit 6695ad2 into main Aug 13, 2026
4 checks passed
@kgraph57
kgraph57 deleted the feat/skill-verification-loops branch August 13, 2026 05:54
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant